TW201928724A - System and method for automated bidding using deep neural language models - Google Patents

Abstract

Systems, devices, and methods are disclosed for predicting potential effectiveness of query-triggered internet advertisements received from different web page publishers using a deep learning neural network language model for clustering queries, and for automatically adjusting bids for advertisements by advertisers based on the predicted potential effectiveness. Using query-clusters rather than queries for adjusting bids for advertisements allows for more accurate and more consistent bidding strategy despite of sparsity in historical advertisement performance data, higher return on investments for the advertisers, and higher revenue for the publishers of the advertisements.

Description

System and method for using deep neural language model in automatic bidding

no

In a network system, a web server can host a web page, which includes a web page with web content and a search tool linked to a search engine to search the web content. A user communication device running a web browser application can directly access the web server To render a web page within a web browser application, the user can also access the web page after redirecting to the web page through a link included on another web page.

In addition to web content and search engines, some web pages are designed to include dedicated display areas or spaces for displaying information in the form of, for example, advertisements. Internet advertising has created new places, new environments and new places for advertisers Platform to show more targeted and cost-effective advertising campaigns to Internet users who may not be exposed to other forms of advertising outside the Internet.

no

In the following, the subject matter of the present invention will be described more fully with reference to the accompanying drawings, which form a part and show specific examples by way of illustration. However, the subject matter of the present invention may be embodied in various forms, and therefore the subject matter covered or claimed is to be construed as, and is not limited to, any illustrated specific embodiments set forth in the text; the illustrated specific embodiments are for illustration purposes only provide. Similarly, what is to be covered is a reasonably broad scope of the claimed or covered subject matter. In addition, for example, the subject matter of the present invention may be a method, an apparatus, a component, or a system. Thus, specific embodiments may, for example, have hardware, software, firmware, or any combination thereof (other than the software itself). Therefore, the following detailed description is not intended as a limiting concept.

Throughout the specification and claims, terminology may have nuances that are taught or implied in context, and that exceed the meaning of a clear statement. Likewise, as used herein, the term "in one embodiment" does not need to refer to the same specific embodiment, and the term "in another embodiment" does not need to refer to a different specific embodiment. For example, the claimed subject matter is intended to include a combination of illustrated specific embodiments in whole or in part.

In general terms are understood at least in part from their use in context. For example, terms such as "and", "or" or "and / or" as used herein may include a variety of meanings based at least in part on the context in which they are used. In general, if you use "or" to associate a list, such as A, B, or C, it is intended to mean the inclusive concepts A, B, and C, and the exclusive concepts A, B, or C, as used herein. In addition, the term "one or more" as used herein is used to describe any feature, structure, or characteristic having a singular concept, or a feature, structure, or combination of characteristics, at least in part, depending on the context. Similarly, terms such as "a" or "an" are also to be understood to convey singular or plural usages at least in part depending on the context. In addition, the term "based on" can be understood as not necessarily conveying a set of exclusive factors, but rather allowing other factors that are not necessarily explicitly described depending at least in part on the context.

It is expected that the description of the specific embodiments described herein can provide a general understanding of the structure of the various specific embodiments. These descriptions are not intended to provide a complete description of all elements and features of a device or system using the structures or methods described herein. After examining the present disclosure, many other specific embodiments will become apparent to those skilled in the relevant art. Many other specific embodiments can be used and derived from the disclosure of the present invention, and therefore, structural and logical substitutions and changes can be made without departing from the scope of the present disclosure. In addition, such descriptions are representative only and are not necessarily drawn to the correct scale. Some parts of the description may be emphasized, while others may be minimized. Accordingly, the disclosure and the drawings are to be regarded as illustrative rather than restrictive.

With the increasing popularity of the Internet and mobile electronic devices, consumers are increasingly relying on web-based content to obtain information. Therefore, a new Internet-based advertising network system was created using previously unavailable Internet technologies. Within an internet-based advertising network system, advertisers are understood to generate and otherwise provide internet advertisements that provide, for example, goods or services for sale. For example, Internet advertising may include audiovisual content, including, but not limited to, text, still images, audio, and / or animation / video. Advertisers can create, upload, and manage their ad content to the ad content database, and control the rendering of ads on web pages such as publishers (or news publishers as described below). Advertisers can maintain their own web pages for their goods and services (or have their web pages hosted). The ads displayed on the publisher's webpage can be hyperlinked to the advertiser's webpage related to the ad.

A publisher is understood as a web page that receives and displays advertisers' internet advertisements in addition to hosting the web page content. For example, publishers can display web content in their web pages while leaving space in the form of one or more slots on the web pages to display online ads. The publisher may also link and control an ad server that selects Internet advertisements to be displayed on the advertising space of the publisher's webpage based on the search query of the user entered into the search engine on the publisher's webpage. Advertisers may agree to pay a cost-per-click (CPC) rate each time a click occurs on an Internet advertisement published on a publisher's webpage. Clicking on an internet ad directs users to the advertiser's webpage.

Publishers can further distinguish between self-owned and operated publishers (hereinafter referred to as OO publishers) and third-party publishers. The OO publisher can control a web server that hosts the main web content provider (for example, Yahoo) and the associated web server. Web servers associated with OO publishers can be specifically identified by advertisers to display their Internet ads. On the other hand, third-party publishers may not be directly associated with OO publishers or advertisers. Instead, third-party publishers can include search engines (eg, Yahoo Search) associated with OO publishers and have an agreement with OO publishers to display the Internet from the same Internet advertising database available to OO publishers Advertising, but may not be affiliated or controlled by the OO publisher. Even so, the third-party publisher page may select the same Internet advertisement as the OO publisher page based on the search query entered into the same search engine. OO publishers can then split CPC revenue with third-party publishers to get clicks on Internet ads displayed on the third-party publisher's web pages.

Publishers can go one step further by sending electronic invitations to ad servers or other parties to allow bidding on their web pages by user requests (eg, by a user typing a search query into the publisher's web page). For example, publishers can control the ad server to manage bids from various advertisers and select ads from advertisers who are willing to pay the highest CPC rate, or they can follow the second price auction model, so the highest bid advertisers pay publishers The second highest bid CPC rate to show its ads.

The ad server can host the bidding platform accordingly. Advertisers can provide ad servers with general guidelines about the CPC rate at which advertisers are willing to accept ads. As will be described in more detail below, an advertiser may further pre-authorize an ad server to adjust bids for displaying ads in a publisher's webpage based on the potential effectiveness of the ads, further generally or specifically. For example, an ad server can estimate the potential effectiveness of an ad based on a set of Internet-centric dimensional information, such as the attributes of the publisher, the attributes of the user device linked to the publisher, the attributes of the advertiser, and the user device The publisher's web page is entered into the publisher's search engine and triggers the ad server to identify the attributes of the query for the ad to be displayed. The ad server may perform this estimation using historical performance records, which are advertisements on various publishers generated by various queries from various user devices. Therefore, the ad server may be further configured to maintain these historical performance records. According to some embodiments, the ad server may be part of a web server for a particular publisher, and according to other embodiments, the ad server may be a separate device that communicates with a web server.

The historical performance of an ad can be determined, for example, by whether it is converted after a user clicks. In particular, by clicking on an advertiser's Internet ad, users can be directed to the advertiser's own webpage that promotes the advertiser's goods or services. Once the user is directed to the advertiser's webpage after clicking, the conversion provided by the advertiser will be completed by the user. Conversion can be understood as a user clicking on one or more of Internet advertisements, a predetermined length of time when a user browses an advertiser's webpage, and a user entering personal information (e.g., name, mailing address) via information entry fields on the advertiser's webpage , Phone number, or email) through the information input field on the advertiser's webpage, or the user actually purchases the goods or services provided by the advertiser on the advertiser's webpage. This conversion can be monitored and recorded by the ad server along with other attributes associated with the ad, such as the attributes of the publisher, advertiser, search query that caused the ad to appear, and the user device that made the query. Analyze a set of ads with the same or similar attributes (dimension information and advertiser information) to provide conversion rates. The conversion rate can be further standardized to a similar reference conversion rate to obtain a conversion rate ratio (hereinafter referred to as CRR). A record CRR with the aforementioned attributes provides a historical performance record of Internet advertising and can be stored in a historical performance database. The source of the CRR is described in more detail in US Patent Application No. 15 / 064,310 by the same applicant, which is incorporated herein by reference in its entirety.

However, due to the sparseness of historical advertising records with the same attributes described above, there may not be a sufficient number of records for a particular combination of attributes to present any statistically significant prediction of the potential performance of an advertisement. In particular, the search query attribute is a huge dimension (with a large number of different search queries), so it has the greatest impact on the issue of data sparsity. As a result, queries associated with ads may not be found in historical performance records.

This disclosure solves the problem by using data aggregation for historical performance records based on at least search query attributes. Specifically, the search query may cluster context and semantic meaning based on deep learning technology, for example, a deep learning neural network language model representing search queries in a dense vector space. In this way, search queries that are close to each other in a dense vector space can be classified into the same search query cluster and are basically the same search. When estimating the potential effectiveness of an advertisement with attributes that include a search query, aggregate historical performance records based on the same search query cluster instead of the same search query. Therefore, the potential effectiveness of significantly more ad traffic processed by the ad server can be estimated with reasonable accuracy, and bid adjustments can be provided to a significantly larger portion of the ad traffic.

The first figure illustrates an exemplary system diagram of a network system 100 for providing bid adjustments for Internet advertisements based on at least dimensional information of ad traffic and deep learning technology. The network system 100 includes an advertisement server 110, a user device 150, an advertiser server 130, an advertisement content repository 134, a web server 120 and 160, a web content repository 125 and 165, and a historical advertisement, which communicate with each other via the network 140. A performance database 181 and a deep learning neural network model 170.

The advertiser server 130 may include a processor 131, a memory 132, and a network interface 133 for accessing the network 140. The advertiser server 130 may generate an Internet advertisement of the advertiser. The Internet advertisement of the advertiser includes advertisement content (for example, audiovisual content) and a link to the advertiser's webpage activated when the user clicks the Internet advertisement (for example, hyper link). Internet advertisements may be stored in an ad content repository 134 in communication with the advertiser server 130. The advertiser server 130 may also control the network interface 133 to send Internet advertisements to other devices and components within the network system 100 via the network.

The web server 120 of the web system 100 hosts one or more web pages related to the OO publisher. The web server 120 may include a processor 121, a memory 122, a search engine 123, and a network interface 124 for connecting to the network 140. The web page content included in the web page hosted by the web server 120 may be stored in a web content repository 125 that communicates with the web server 120. The search engine 123 may receive a user's search query and search the web content repository 125 for related web content. The web server 120 may select web content from the web content repository 125 based on the search results determined by the search engine 123 to be displayed on the host web page. The web server 120 may also receive internet advertisements from the ad server 110 for presentation on a hosted web page.

The publisher web server 160 of the web system 100 may include a third-party web server 160 related to one or more web pages hosted by a third-party publisher. The third-party web server 160 may include a processor 161, a memory 162, a search engine 163, and a network interface 164 for connecting to the network 140. The webpage content included in the webpage hosted by the third-party web server 160 may be stored in a webpage content repository 165 in communication with the third-party web server 160. The search engine 163 may receive a user's search query and search the web content repository 165 for related web content. The third-party web server 160 may select web content from the web content repository 165 to display on the hosted web page based on the search results determined by the search engine 163. The third-party web server 160 may also receive internet advertisements from the ad server 110 for presentation on the hosted web page. According to an embodiment, the search engine 163 may be connected to the search engine 123 to return the search results to the search query input into the search engine 163 and the search engine 123 using the same logic.

The communication device 150 in the network system 100 may include a processor 151, a memory 152, and a network interface 153 for connecting to the network 140. The communication device 150 may be, for example, a desktop computer, or a mobile phone, a smart phone, a display pager, a radio frequency (RF) device, an infrared (IR) device, a personal digital assistant (PDA), a handheld computer, or a tablet computer. , Notebook computers, desktop boxes, wearable computers, portable devices that combine, for example, integrated devices that combine various features of the aforementioned devices. In addition, the communication device 150 may include or execute various possible applications, such as a client software application capable of communicating with other devices by transmitting one or more information. Several possible examples may be, for example, email, short-term Messaging services (SMS) or multimedia messaging services (MMS), including through networks such as social networks, including, for example: Facebook, LinkedIn, Twitter, Flickr or Google+. The communication device 150 may further include or execute an application program for communication content such as text content, multimedia content, and the like. The communication device 150 may also include or perform applications for various possible tasks, such as browsing, searching, playing various forms of content or games (such as fantasy sports leagues) including locally stored or streaming videos. The foregoing is intended to illustrate that the subject matter of the invention is extensible to include a wide range of possible features or capabilities.

For example, the communication device 150 may run a web browser application for connecting to a web page hosted by the web server 120 or a third-party web server 160 and presenting web page content related to the hosted web page within the web browser application. The second figure illustrates an exemplary web browser 200 running on, for example, a communication device 150, and presenting a web page hosted by, for example, a web server 120 or a third-party web server 160. The web page presented by the web browser 200 includes a search tool 201, where the search tool 201 may correspond to the search engine 123 or the search engine 163 included in the web server 120 or the third-party web server 160, respectively. The webpage further includes a first sponsored advertisement list 202, a fixed position advertisement 204, and a second sponsored advertisement list 205, and each of the first sponsored advertisement list 202, the fixed position advertisement 204, and the second sponsored advertisement list 205 can be set as biddable. The advertisement may be adjusted and selected by, for example, a bid handler 113 and an advertisement selector 114 of the advertisement server 110 described below based on search input and / or search results analyzed by the search engine 123 or the search engine 163. The web page presented by the web browser 200 may further include web content results 203, which are respectively generated by, for example, the search engine 123 or the search engine 163 based on a search query entered by the user in the search tool 201.

Returning to the first figure, the advertisement server 110 of the network system 100 may include a processor 111, a memory 112, a bid processor 113, an advertisement selector 114, an advertisement performance monitor 116, and a network for connecting to the network 140. Interface 115. When a user performs a search (via, for example, a bid invitation), the ad server 110 may obtain the dimension information from the web server 120 or 160. The ad server 110 may receive a bid advertisement from an advertiser, and select a winding advertisement to be displayed on a webpage hosted by the web server 120 or 160 based on the received dimensional information and bid advertiser information.

Specifically, the bid processor 113 of the ad server 110 may determine whether the potential effectiveness of the bid ad predicted using the historical ad performance record, the dimensional information received from the web server 120 or the publisher web server 160, and the advertiser information determine whether Adjust bids from advertisers. The dimensional information may include, for example, search query information (such as a search query of a user entered into the search engine 123 or the search engine 163), communication device attribute information (such as information identifying a device type of the communication device 150), and / or publisher identification information (Such as identifying the web server 120 or the third-party web server 160 as information associated with an OO publisher or a third-party publisher or a specific publisher).

The historical ad performance record may be maintained in a historical ad performance database 181 in communication with the network 140 and linked by the bid processor 113 of the ad server 110. Maintain historical ad performance records for tracking displayed ad performance with a specific combination of dimension information and advertiser information. Recorded performance may include performance metrics such as conversion rate (CRR). For example, historical performance records may provide specific CRR for advertising history with a specific combination of dimensional information. Therefore, the ad server 110 may include an advertising performance monitor 116 to track historical advertising. record of. In some embodiments, the historical advertisement performance database 181 may be part of the advertisement server 110. In some other embodiments, a dedicated server may be used to perform ad performance monitoring.

The third figure illustrates an exemplary collection of historical performance records 300, which includes records 302, 304, 306, 308, 310, 312, and 313. Each record includes a data field of performance index CRR 320 and includes a search query 332, device attributes 334. Dimension information 330 of the data fields of the publisher information 336 and the advertiser information 338. In an embodiment, when historical CRR is unavailable or not statistically significant for a performance record for a particular combination of dimensions such as record 306, the CRR field may be labeled as "N / A", for example, if relevant If the number of historical ad clicks related to the performance record is less than the predetermined number, such as 500, the historical CRR can be considered statistically irrelevant. Therefore, due to the large number of combinations of dimensional information, it may not correspond to statistically significant CRR. Can be sparse.

Returning to the first figure, the bid processor 113 of the ad server 110 identifies records having the same or similar dimensional information as bid advertisements or some combination of them from the historical advertisement performance database, and uses the identified history The corresponding performance indicators of the advertising records are predicted to obtain a prediction of the potential effectiveness of the bid advertisements. For example, when the bid processor 113 is ready to determine a validity prediction for the current click traffic received by the search query A, received from the publisher web page B, and linked from the communication device type C to the publisher web page B, the bid processor 113 may identify that it has Records in the historical advertisement performance database 181 (eg, 300 in the third figure) of the same or similar search query A, publisher webpage B, and communication device type C. The bid processor 113 may then predict the potential effectiveness of the bid advertisement based on the identified records and corresponding historical performance indicators. The query cluster function of the deep learning neural network model 170 that may be involved in this prediction process will be described in more detail as follows.

According to some embodiments, the ad server 110 may be included as part of the web server 120, while in other embodiments, the ad server 110 may be a stand-alone computing device as shown in the first figure. The ad server 110 may also communicate with an ad content database 134 that stores Internet ads. Internet advertisements stored in the advertisement content database 134 may have been received by the network 140 by the network interface 115 of the advertiser server 130.

The web content and Internet advertising content described by the web server 120, the third-party web server 160, and the ad server 110 may be integrated together before being transmitted to a web browser application run by the communication device 150.

The neural network model 170 of the first figure may be implemented by multiple server computers. For example, the neural network 170 may be implemented by server computers 171-173. Neural network 170 may be configured to apply deep learning techniques (eg, query2vec and / or word2vec) to help incorporate known queries received by web search engines (eg, search engine 123 and / or search engine 163) into a query cluster in. For example, deep neural network language models 170 can be trained using historical data powered by computers 171-173 and trained to learn high-quality vector representations of user search queries, and these representations can be used by standard clustering techniques (for example: k-average , Hierarchical clustering, etc.) to group similar user search queries into a predetermined number of query clusters. The clustering strategy implemented by the deep learning neural network language model 170 embodiment may consider creating a predetermined number of query clusters, which optimizes the robustness of the search query within each query cluster. In order to satisfy the creation of a predetermined number of query clusters, the neural network 170 may set the user search query similarity sensitivity level to group the queries into a predetermined number of query clusters (eg, 1000 query clusters).

Then, the historical performance record 300 of the third graph can be viewed based on the query cluster instead of a separate query. In an embodiment, as shown in the fourth figure, the query cluster information 410 may be replaced with the search query information 332 in the third figure to obtain the modification history performance record 400. Deep learning neural language model 170 self-learns high-quality vector representations of user search queries and groups semantically similar queries based on the content of the words in the query together, for example, because "women's shoes" and "sneakers" are considered to be Refers to similar items in the vector representation, and "women's shoes" and "Chanel perfume" are considered to be related to women, so the queries in records 302, 306, 310, and 312 can be grouped into one group (query cluster 1).

The deep learning neural network language model 170 may also be used to determine a query cluster for a bid advertisement. Even when the historical performance record 300 of the third graph is sparse relative to the specific search query associated with the bid advertisement, the bid processor 113 of the ad server 110 of the first graph may use the obtained query cluster in conjunction with the fourth graph In the modification history performance record 400 in, obtain the effectiveness prediction of the bid advertisement.

The fifth figure illustrates an exemplary logic flow that the bid processor 113 and the ad selector 114 of the ad server 110 can follow. The ad server may receive dimensional information and advertiser information for the current ad (502). The ad server may link with the historical performance database 181 to identify historical performance records with the same or similar dimensional information and advertiser information as the current bid advertisement (504). If such a record is identified, the ad server then determines whether historical performance indicators are available for the identified historical performance record (506). If historical performance metrics are available, the ad server can then use the historical performance metrics to adjust the bid for the current bid ad (508). If no record in the historical performance record is identified as having the same or similar dimensional information and advertiser information as the bid advertisement ("No" branch of 504) or such a record is identified but the related historical performance indicator is not available ("No" in 506 ”Branch), the ad server can use the deep learning neural network model 170 to obtain a query cluster of current bid advertisements (510), and identify records from the modified historical performance record 400 instead of the original performance record 300 to predict the current bid advertisement ’s Potential effectiveness (512). In one embodiment, the ad server can identify a set of records in the modification history performance record 400 having the same query cluster, the same or similar other dimension information, and the same advertiser information (512). The ad server may further aggregate the available performance metrics that identify the records (in 512) to obtain aggregated historical performance metrics (514). The ad server then adjusts bids on bidding ads based on aggregated historical performance metrics (516).

In the ad server, the above terms "same or similar" may be flexibly defined. For example, the ad server may additionally use techniques such as fuzzy text matching to determine whether the queries are the same or similar. Therefore, differences between queries such as misspellings may not cause the queries to be considered different or similar.

In one embodiment, the ad server may not execute the "yes" branch of 504, so that even if there is a historical performance record that exactly matches the current bid suggestion, the prediction of the potential effectiveness and bid adjustment is based only on the query cluster. This method uses more historical data to make predictions, so it can provide better prediction accuracy in some cases.

In addition, a hierarchical approach can be used to predict the potential effectiveness of bidding advertisements (512 and 514). For example, one of the dimensions information and advertiser information may be omitted. Specifically, the advertiser information, device information, publisher information, or any combination thereof may be omitted in steps 512 and 514. The further aggregation of historical performance indicator information based on these dimensions will make more historical data available for prediction. Similarly, some dimensional information and advertiser information can be omitted in step 504 of the logic flow. Therefore, the "yes" branch of 504 can be implemented in a similar hierarchical manner.

An example is shown using the historical performance record 300 of the third figure and the modified history performance record 400 of the fourth figure. Assume that the current bid advertisement has the following attributes: search query "neckless", device "mobile phone", publisher A, advertiser A, and the ad server 110 identifies no exact or similar between the current bid advertisement and the historical performance record 300 Match. Next, the ad server uses a deep learning neural network language model to determine that the search query "neckless" of the current bid ad belongs to query cluster 1. The ad server then links to modify historical performance record 400 and identifies records 302 and 306 as being the same or similar to the current bid advertisement. Therefore, records 302 and 306 are aggregated to obtain an aggregate CRR of 1.1 (the CRR of record 302 is 1.1 and record 306 is N / A). In the above-mentioned hierarchical implementation, advertiser information can be omitted, and therefore, record 310 is included in the data aggregation, and in this case, the aggregate CRR can be 1.05 (for example, the average of records 302 and 310).

Aggregated CRR can be used to adjust bids. For example, if the aggregated CRR is 1.3, advertisers' initial bids can increase by 30%. The adjusted bid can then serve as the basis for the ad selector to select ads from other bidding advertisers. The permission to adjust bids may be pre-authorized by the advertiser or may be provided on-the-fly by the advertiser ’s server, so the advertiser ’s server may implement an automatic evaluation algorithm to analyze the proposed bid adjustment and allow or deny the bid adjustment in real time. The ad server can continue to use advertiser-provided bidding ads that do not make suggested bid adjustments using the logic in the fifth graph (if there are no matching records in 504 and 512, or there are no performance indicators in the records identified in 504 and 512) Of the original bid value of.

A solution that uses deep learning to group queries into query clusters helps generate more accurate bid adjustment recommendations for more search queries and significantly higher traffic segments. In addition, using deep learning neural network language models instead of other natural language processing models for query clusters helps to extract complex linguistic, semantic, and contextual relationships to obtain better query clusters and more accurate and consistent bid adjustments. More accurate and higher bids can lead to higher conversion rates and higher return on investment (ROI) for advertisers, and higher revenue for publishers.

Returning to the first figure, the network 140 of the network system 100 may include any combination of long distance and access networks. For example, the network 140 may include a wireless network configured to couple the communication device 150 to the Other client devices coupled to the wireless network. The wireless network can be used in stand-alone ad-hoc networks, mesh networks, wireless LAN (WLAN) networks, and cellular networks. The wireless network may further include a system of terminals, gateways, routers, etc. coupled by a radio frequency link, etc., so the network topology may change over time, or even quickly. Wireless networks can further apply multiple network access technologies, including Long Term Evolution (LTE), WLAN, Wireless Routing (WR) networks, or second, third, or fourth generation (2G, 3G, or 4G) Honeycomb technology, etc. Network access technology can generate wide area coverage for devices such as client devices with varying levels of mobility. For example, the network may allow RF or wireless communication via one or more network access technologies, such as Global System for Mobile Communications (GSM), Universal Mobile Telecommunications System (UMTS), Universal Packet Radio Service (GPRS), Enhanced Data Global System for Mobile Communications (EDGE), 3GPP Long Term Evolution (LTE), Advanced LTE, Wideband Code Division Multiple Access (WCDMA), Bluetooth, Bluetooth Low Energy Technology (BLE) as part of Bluetooth Core Specification Version 4.0 ), 802.11b / g / n, etc. A wireless network includes virtually any type of wireless communication mechanism through which signals are transmitted between devices such as client devices or computing devices, between networks 140 or within, and the like.

Via the network 140, which includes a network participating in a digital communication network, the transmitted signal packets are compatible or consistent with one or more protocols. The signaling formats or protocols used include: TCP / IP, UDP, DECnet, NetBEUI, IPX, APPALETALKTM, etc. Internet Protocol (IP) versions can include IPv4 or IPv6. The Internet is a decentralized global network of networks. The Internet includes local area networks (LANs), wide area networks (WANs), wireless networks, or long-haul public networks. For example, it enables signal packets to be transmitted between LANs. The signal packet is transmitted between 140 nodes in the network, for example, to one or more addresses of an application area network address. For example, a signal packet may be transmitted over the Internet from a user address via an access node coupled to the Internet. Similarly, for example, a signal packet may be forwarded through a network node to a destination address coupled to the network 140 through a network access node. For example, a signal packet transmitted via the Internet is routed via a path of a gateway, a server, etc., which routes the signal packet according to the destination address and the availability of a network path to the destination address.

Network 140 may also include a "content delivery network" or "content decentralized network" (CDN), commonly referred to as a decentralized content delivery system, which includes a collection of computers or computing devices connected using a network. CDNs can use software, systems, protocols, or technologies to facilitate many services, such as storage, caching, content communications, or streaming media or applications. Services can also utilize supporting technologies, including but not limited to "cloud computing", decentralized storage, DNS request handling, provisioning, signal monitoring and reporting, content targeting, personalization, or business intelligence. The CDN also allows one entity to fully or partially operate or manage another site infrastructure.

The network 140 may also include a peer-to-peer (or P2P) network, where a P2P network may employ the computing power or bandwidth of network participants, as opposed to a network that may use a dedicated device such as a dedicated server. However, some networks can use both and other methods. P2P networks are often used to couple nodes through ad hoc placement or configuration. A peer-to-peer network can use nodes that can operate as "clients" and "servers."

Regarding the network 140, the network 140 may couple devices for communication, such as between the web server 120, third-party web server 160 and communication device 150, or other devices within the network system 100, including, for example, by wireless network Road coupling between wireless devices. The network 140 may also include a large number of storage devices, such as a network attached storage device (NAS), a storage area network (SAN), or other forms of computer or machine-readable media. The network 140 may include the Internet, one or more local area networks (LAN), one or more wide area networks (WAN), wired connections, wireless connections, or any combination of these. Similarly, sub-networks that can use different architectures or can conform to or be compatible with different communication protocols operate within the larger network 140. A variety of devices may be used, for example, to provide interoperability to different architectures or protocols. For one example, routers can provide individual and independent LAN connections. The communication link or channel may include, for example, analog telephone lines, such as twisted pair, coaxial cable, all or part of the digital line, including T1, T2, T3 or T4 type lines, overall service digital network (ISDN), digital subscriber line (DSL), wireless links containing satellite links, or other communication links or channels, as may be known to those skilled in the art. Furthermore, a computer device or other related electronic device may be remotely coupled to the network 140, such as via a telephone line or link.

The sixth figure illustrates an exemplary computer architecture of the computer 600. Embodiments of the computer 600, including additional components and embodiments including fewer components than described, may represent any one or more devices including the network system 100 as shown in the first figure.

The computer 600 includes a network interface device 620 that allows communication with other computers via a network 626, where the network 626 can be represented as the network 140 in the first figure. The computer 600 may include a processor 602, a main memory 604, a static memory 606, a network interface device 620, an output device 610 (for example, a display or a speaker), an input device 612, and a storage device 616, all of which are connected by a bus 608 .

The processor 602 represents a central processing unit of any type of architecture, such as CISC (Complex Instruction Set Computing), RISC (Reduced Instruction Set Computing), VLIW (Very Long Instruction Word), or mixed architecture, and other suitable processing may be used Device. The processor 602 executes instructions and includes portions of the computer 600 that control the operation of the entire computer 600. The processor 602 may also represent a controller that organizes data and program storage in memory and transfers data and other information between various parts of the computer 600.

The processor 602 is configured to receive input data and / or user commands from the input device 612. The input device 612 may be a keyboard, a mouse or other pointing device, a trackball, a scroll, a button, a touchpad, a touch screen, a keyboard microphone, a voice recognition device, a video recognition device, or any other suitable mechanism for a user Data is entered into the computer 600 and controls the operation of the computer 600 and / or the operation of the processing steps and other features described herein. Although only one input device 612 is shown, in another embodiment, any number and type of input devices may be included, for example, the input device 612 may include an accelerometer, a gyroscope, and a global positioning system (GPS) transceiver.

The processor 602 can also communicate with other computers via the network 626 to receive instructions 624. The processor can control the instructions 624 to be stored in a main memory 604 such as a random access memory (RAM), such as a read-only memory (ROM) ) In any one or more of the static memory 606 and the storage device 616. The processor 602 may then read and execute instructions 624 from any one or more of the main memory 604, the static memory 606, or the storage device 616. The instructions 624 may also be stored on any one or more of the main memory 604, the static memory 606, or the storage device 616 by other sources. The instruction 624 may correspond to, for example, an instruction representing the above-mentioned bid processor 113.

Although the computer 600 is shown to include only a single processor 602 and a single bus 608, the disclosed embodiments are equally applicable to computers that may have multiple processors and computers that may have multiple buses, some or all of which are different Way to perform different functions

The storage device 616 represents one or more mechanisms for storing data. For example, the storage device 616 may include computer-readable media 622, such as read-only memory (ROM), RAM, non-volatile storage media, optical storage media, flash memory devices, and / or other machine-readable media. media. In other embodiments, any suitable type of storage device may be used. Although only one storage device 616 is shown, there may be multiple storage devices and multiple types of storage devices. Further, although the computer 600 is shown as including a storage device 616, it may be distributed on other computers, for example, on a server.

The storage device 616 may include a controller (not shown) and a computer-readable medium 622 having instructions 624 that can be executed by the processor 602 to perform the methods described above. In another embodiment, some or all functions are performed by hardware instead of a processor-based system. In one embodiment, the controller is a web browser, but in other embodiments, the controller may be a database system, a file system, an email system, a media manager, an image manager, or may include a data item capable of linking data items. Any other function. The storage device 616 may also include additional software and data (not shown) that are not necessary to understand the features described above.

The output device 610 is configured to present information to a user. For example, the output device 610 may be a display, such as a liquid crystal display (LCD), a gas or plasma-based flat panel display, or a traditional cathode ray tube (CRT) display or other display device. A type of display known in the field of computer hardware. Therefore, in some embodiments, the output device 610 displays a user interface. In other embodiments, the output device 610 may be configured as a speaker that outputs audible information to a user. In other embodiments, any combination of output devices may be represented by the output device 610.

The network interface device 620 provides the computer 600 to connect to the network 626 via any suitable communication protocol. The network interface device 620 sends and / or receives data from the network 626 via a wireless or wired transceiver 614. The transceiver 614 may be a cellular frequency, radio frequency (RF), infrared (IR), or many known wireless or wired transmission system capable of communicating via the network 626 or other computer devices having some or all of the characteristics of the computer 600. any type. The bus 208 may represent one or more buses, for example, USB, PCI, ISA (Industrial Standard Architecture), X-bus, EISA (Extended Industrial Standard Architecture), or any other suitable bus and / or bridge (Also known as a bus controller).

The computer 600 may be implemented using any suitable hardware and / or software, such as a personal computer or other electronic computing device. In addition to the aforementioned various types of wearable devices, the computer 600 may be a portable computer, a laptop computer, a tablet computer or a notebook computer, a PDA, a pocket computer, an electrical appliance, a telephone, or a host computer.

The disclosed method and system can be implemented partly in a server, a client device, a cloud computing environment, partly in a server and partly in a client device, or a combination of a server, a cloud computing environment, and a client device.

Therefore, the foregoing detailed description should be regarded as illustrative rather than limiting, and it should be understood that the scope of patent application (including all equivalents) described below is used to define the spirit and scope of the present invention.

100‧‧‧ Network System

110‧‧‧Ad Server

111‧‧‧ processor

112‧‧‧Memory

113‧‧‧Bid Processor

114‧‧‧Ad Selector

115‧‧‧ network interface

116‧‧‧ Advertising Performance Monitor

120‧‧‧Web Server

121‧‧‧ processor

122‧‧‧Memory

123‧‧‧Search Engine

124‧‧‧Interface

125‧‧‧Web Content Repository

130‧‧‧Advertiser server

131‧‧‧ processor

132‧‧‧Memory

133‧‧‧Interface

134‧‧‧Ad Content Repository

140‧‧‧Internet

150‧‧‧user device

151‧‧‧Processor

152‧‧‧Memory

153‧‧‧Interface

160‧‧‧Web Server

161‧‧‧Processor

162‧‧‧Memory

163‧‧‧Search Engine

164‧‧‧Interface

165‧‧‧Web Content Repository

170‧‧‧ Deep Learning Neural Network Model

171‧‧‧Server Computer

172‧‧‧Server Computer

173‧‧‧Server Computer

181‧‧‧History Advertising Performance Database

200‧‧‧ web browser

201‧‧‧Search tools

202‧‧‧First Sponsored Advertising List

203‧‧‧Web Content Results

204‧‧‧Fixed ad

205‧‧‧Second Sponsored Advertising List

300‧‧‧Historical Performance Record

302‧‧‧Record

304‧‧‧Record

306‧‧‧Record

308‧‧‧Record

310‧‧‧Record

312‧‧‧Record

313‧‧‧Record

320‧‧‧ Performance Index

330‧‧‧ Dimensional Information

332‧‧‧Search query

334‧‧‧Device Properties

336‧‧‧Poster Information

338‧‧‧ Advertiser Information

400‧‧‧ modify historical performance records

410‧‧‧Query cluster information

502-516‧‧‧step

600‧‧‧ computer

602‧‧‧ processor

604‧‧‧Main memory

606‧‧‧Static memory

610‧‧‧Output device

612‧‧‧input device

614‧‧‧ Transceiver

616‧‧‧Storage device

620‧‧‧Network Interface Device

622‧‧‧Computer-readable media

624‧‧‧Instruction

626‧‧‧Internet

The first figure illustrates a system diagram of an exemplary advertising network system.

The second figure illustrates an exemplary web browser interface displaying a web page.

The third figure illustrates an example of historical performance records for web advertising.

The fourth figure illustrates an exemplary modification of the historical performance records of Internet advertisements based on a query cluster.

The fifth figure illustrates an exemplary logical flow for realizing the prediction of the potential effectiveness of an advertisement and providing ad bid recommendations.

The sixth diagram illustrates a block diagram of an exemplary computer architecture of one or more components included in the exemplary advertising network system shown in the first diagram.

Claims (20)

A server including: A network interface configured as: Receiving a bid invitation for advertising space from an information publishing server based on a query received from the requester at the aforementioned information publishing server; and Receiving a bid from an advertiser in response to the aforementioned query and the aforementioned bid invitation; and A processor that communicates with the aforementioned network interface is configured as: Use a training query cluster neural network to determine a query cluster for the aforementioned query in a set of predefined query clusters; Aggregate historical performance records of query-based advertisements according to at least query clusters determined by the aforementioned training query cluster neural network to obtain aggregated historical performance records; Predicting an effectiveness measure of the advertising space used for the aforementioned queries of the aforementioned advertisers based on the aforementioned query clusters and the aforementioned aggregated historical performance records; and An adjustment to the bid received from the advertiser is automatically controlled according to the forecast effectiveness measure to obtain an adjusted bid. The server according to item 1 of the scope of patent application, wherein the historical performance records of the aforementioned query-based advertisements include: A query field, which specifies the query corresponding to the aforementioned query-based advertisement; A device type field, which clearly indicates the type of query device corresponding to the aforementioned query-based advertisement; A publisher field identifying the publisher of the aforementioned query-based advertisement; An advertiser field identifying an advertiser corresponding to the aforementioned query-based advertisement; and A performance indicator field, which clearly lists the performance indicators corresponding to the aforementioned query-based advertisement. The server according to item 2 of the scope of patent application, wherein the performance index of the query-based advertisement corresponds to a conversion rate ratio between a conversion rate of the query-based advertisement and a reference conversion rate. The server described in item 3 of the scope of patent application, Wherein, the aforementioned conversion corresponds to a querying device browsing a webpage associated with the query-based advertisement for a predetermined length of time, the querying device entering information on the webpage, or a purchase via the querying device is recorded in the aforementioned On the web; and The conversion rate corresponds to the number of conversions of the query-based advertisement and is normalized to the number of views of the query-based advertisement. The server according to item 2 of the scope of patent application, wherein the processor is configured to aggregate the historical performance record in the following manner to obtain the aggregated historical performance record: Converting the aforementioned query field into a query cluster field using the aforementioned training query cluster neural network; and Aggregate performance indicators with historical performance records of the same query cluster field, device type field, publisher field, and advertiser field to obtain the previously aggregated historical performance records. The server according to item 5 of the scope of patent application, wherein the aforementioned processor is configured to predict the aforementioned effectiveness measure of the advertising space regarding the aforementioned query of the aforementioned advertiser by: A record is identified from the aggregated historical performance record, the record has a query cluster field, a publisher field, a device type that match the query cluster, the publishing server, the advertiser, and the requester, respectively, An advertiser field and a device type field; and Based on the aggregated performance index of the aforementioned identification record, the aforementioned effectiveness measure of the advertising space of the aforementioned query of the advertiser is predicted. The server according to item 2 of the scope of patent application, wherein the processor is configured to aggregate the historical performance record to obtain the aggregated historical performance record by: Converting the aforementioned query field into a query cluster field using the aforementioned training query cluster neural network; and Regardless of the aforementioned advertiser field, aggregate performance indicators with historical performance records of the same query cluster field, device type field, and publisher field to obtain the aggregated historical performance record. The server according to item 2 of the scope of patent application, wherein the processor is configured to aggregate the historical performance record to obtain the aggregated historical performance record by: Converting the aforementioned query field into a query cluster field using the aforementioned training query cluster neural network; and Regardless of the aforementioned publisher field, aggregate performance indicators with historical performance records of the same query cluster field, device type field, and advertiser field to obtain the aggregated historical performance record. The server according to item 2 of the scope of patent application, wherein the processor is configured to aggregate the historical performance record to obtain the aggregated historical performance record by: Converting the aforementioned query field into a query cluster field using the aforementioned training query cluster neural network; and Regardless of the aforementioned advertiser field and the aforementioned publisher field, performance indicators with historical performance records having the same query cluster field and device type field are aggregated to obtain the aforementioned aggregated historical performance record. The server according to item 1 of the patent application scope, wherein the aforementioned training query cluster neural network is configured to convert a query into a vector representation. The server as described in claim 10, wherein the training query cluster neural network is further configured to cluster the aforementioned queries based on the aforementioned vector representation of the cluster. The server as described in claim 10 or 11, wherein the aforementioned query includes a query dialog text, and wherein the aforementioned training query cluster neural network is configured to use the context information of words in the aforementioned query dialog text to transform the aforementioned query Conversation text. A method including: Receiving a bid invitation for advertising space from an information publishing server based on a query received from a requester at the aforementioned information publishing server; Receiving a bid from an advertiser in response to the aforementioned query and the aforementioned bid invitation; and Use a training query cluster neural network to determine a query cluster for the aforementioned query in a set of predefined query clusters; Aggregate historical performance records of query-based advertisements according to at least query clusters determined by the aforementioned training query cluster neural network to obtain aggregated historical performance records; Predicting an effectiveness measure of the advertising space used for the aforementioned queries of the aforementioned advertisers based on the aforementioned query clusters and the aforementioned aggregated historical performance records; and According to the aforementioned forecast effectiveness measure, an adjustment to the aforementioned bid received from the aforementioned advertiser is automatically controlled to obtain an adjusted bid. The method according to item 13 of the scope of patent application, wherein the historical performance records of the aforementioned query-based advertisements include: A query field, which specifies the query corresponding to the aforementioned query-based advertisement; A device type field, which clearly indicates the type of query device corresponding to the aforementioned query-based advertisement; A publisher field identifying the publisher of the aforementioned query-based advertisement; An advertiser field identifying an advertiser corresponding to the aforementioned query-based advertisement; and A performance indicator field, which clearly lists the performance indicators corresponding to the aforementioned query-based advertisement. The method according to item 14 of the scope of patent application, wherein the performance index of the query-based advertisement corresponds to a conversion rate ratio between a conversion rate of the query-based advertisement and a reference conversion rate. The method described in the scope of application of the patent, Wherein, the aforementioned conversion corresponds to a querying device browsing a webpage associated with the query-based advertisement for a predetermined length of time, the querying device entering information on the webpage, or a purchase via the querying device is recorded in the aforementioned On the web; and The conversion rate corresponds to the number of conversions of the query-based advertisement and is normalized to the number of views of the query-based advertisement. The method according to item 14 of the scope of patent application, wherein aggregating the aforementioned historical performance records to obtain the aforementioned aggregated historical performance records includes: Converting the aforementioned query field into a query cluster field using the aforementioned training query cluster neural network; and Aggregate performance indicators with historical performance records of the same query cluster field, device type field, publisher field, and advertiser field to obtain the previously aggregated historical performance records. The method according to item 17 of the scope of patent application, wherein the aforementioned effectiveness measure of predicting the advertising space of the aforementioned query of the aforementioned advertiser includes: A record is identified from the aggregated historical performance record, the record has a query cluster field, a publisher field, a device type that match the query cluster, the publishing server, the advertiser, and the requester, respectively, An advertiser field and a device type field; and Based on the aggregated performance index of the aforementioned identification record, the aforementioned effectiveness measure of the advertising space of the aforementioned query of the advertiser is predicted. A server including: A query cluster server configured to implement a training query cluster neural network; and An ad server, including: A database for storing historical performance records of query-based ads; A network interface, configured as: Receiving a bid invitation for advertising space from an information publishing server based on a query received from a requester at the aforementioned information publishing server; Receiving a bid from an advertiser in response to the aforementioned query and the aforementioned bid invitation; Sending the query to the query cluster server to use the trained query cluster neural network to determine a query cluster for the query; and Receiving the aforementioned query cluster of the aforementioned query from the aforementioned query cluster server; and A processor in communication with the aforementioned network interface is configured to: Obtaining the aforementioned historical performance records of query-based advertisements from the aforementioned database; Aggregate the aforementioned historical performance records of the query-based advertisements according to at least the query clusters determined by the aforementioned training cluster neural network of the aforementioned query cluster server to obtain aggregated historical performance records; Predicting an effectiveness measure of the advertising space used for the aforementioned queries of the aforementioned advertisers based on the aforementioned query clusters and the aforementioned aggregated historical performance records; and An adjustment to the bid received from the advertiser is automatically controlled to obtain an adjusted bid according to the aforementioned prediction effectiveness measure. The server as described in item 19 of the scope of patent application, The historical performance records of the aforementioned query-based advertisements include: a query field that explicitly lists the queries corresponding to the aforementioned query-based ads; a performance indicator field that explicitly lists the corresponding ones Performance indicators; and at least one third field; and The processor is configured to aggregate the historical performance records in the following manner to obtain the aggregated historical performance records: Converting the aforementioned query field into a query cluster field using the aforementioned training query cluster neural network; and Aggregate performance indicators with historical performance records of the same query cluster field and third field to obtain the previously aggregated historical performance records.